From cd548836a1410220090dbd7866735a7493c7cf47 Mon Sep 17 00:00:00 2001 From: James Benton Date: Wed, 30 May 2012 14:36:22 +0100 Subject: gallivm: Add support for unorm16 in lp_build_mul. Reviewed-by: Jose Fonseca --- src/gallium/auxiliary/gallivm/lp_bld_arit.c | 45 +++++++++++++++++++++++++++++ 1 file changed, 45 insertions(+) (limited to 'src/gallium/auxiliary/gallivm/lp_bld_arit.c') diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c index d23ff0b..ca96a6b 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c @@ -691,6 +691,35 @@ lp_build_mul_u8n(struct gallivm_state *gallivm, return ab; } +/** + * Normalized 16bit multiplication. + * + * Utilises same principle as above code. + */ +static LLVMValueRef +lp_build_mul_u16n(struct gallivm_state *gallivm, + struct lp_type i32_type, + LLVMValueRef a, LLVMValueRef b) +{ + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef c16; + LLVMValueRef ab; + + assert(!i32_type.floating); + assert(lp_check_value(i32_type, a)); + assert(lp_check_value(i32_type, b)); + + c16 = lp_build_const_int_vec(gallivm, i32_type, 16); + + /* ab/65535 ~= (ab + (ab >> 16) + 0x8000) >> 16 */ + ab = LLVMBuildMul(builder, a, b, ""); + ab = LLVMBuildAdd(builder, ab, LLVMBuildLShr(builder, ab, c16, ""), ""); + ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(gallivm, i32_type, 0x8000), ""); + + ab = LLVMBuildLShr(builder, ab, c16, ""); + + return ab; +} /** * Generate a * b @@ -736,6 +765,22 @@ lp_build_mul(struct lp_build_context *bld, return ab; } + if(type.width == 16) { + struct lp_type i32_type = lp_wider_type(type); + LLVMValueRef al, ah, bl, bh, abl, abh, ab; + + lp_build_unpack2(bld->gallivm, type, i32_type, a, &al, &ah); + lp_build_unpack2(bld->gallivm, type, i32_type, b, &bl, &bh); + + /* PMULLW, PSRLW, PADDW */ + abl = lp_build_mul_u16n(bld->gallivm, i32_type, al, bl); + abh = lp_build_mul_u16n(bld->gallivm, i32_type, ah, bh); + + ab = lp_build_pack2(bld->gallivm, i32_type, type, abl, abh); + + return ab; + } + /* FIXME */ assert(0); } -- cgit v1.1